projects/linux·blob v2.6.34 · refs

1850 lines · 45.4 KiB

1/*
2 * linux/kernel/hrtimer.c
3 *
4 * Copyright(C) 2005-2006, Thomas Gleixner <[email protected]>
5 * Copyright(C) 2005-2007, Red Hat, Inc., Ingo Molnar
6 * Copyright(C) 2006-2007 Timesys Corp., Thomas Gleixner
7 *
8 * High-resolution kernel timers
9 *
10 * In contrast to the low-resolution timeout API implemented in
11 * kernel/timer.c, hrtimers provide finer resolution and accuracy
12 * depending on system configuration and capabilities.
13 *
14 * These timers are currently used for:
15 * - itimers
16 * - POSIX timers
17 * - nanosleep
18 * - precise in-kernel timing
19 *
20 * Started by: Thomas Gleixner and Ingo Molnar
21 *
22 * Credits:
23 * based on kernel/timer.c
24 *
25 * Help, testing, suggestions, bugfixes, improvements were
26 * provided by:
27 *
28 * George Anzinger, Andrew Morton, Steven Rostedt, Roman Zippel
29 * et. al.
30 *
31 * For licencing details see kernel-base/COPYING
32 */
33
34#include <linux/cpu.h>
35#include <linux/module.h>
36#include <linux/percpu.h>
37#include <linux/hrtimer.h>
38#include <linux/notifier.h>
39#include <linux/syscalls.h>
40#include <linux/kallsyms.h>
41#include <linux/interrupt.h>
42#include <linux/tick.h>
43#include <linux/seq_file.h>
44#include <linux/err.h>
45#include <linux/debugobjects.h>
46#include <linux/sched.h>
47#include <linux/timer.h>
48
49#include <asm/uaccess.h>
50
51#include <trace/events/timer.h>
52
53/*
54 * The timer bases:
55 *
56 * Note: If we want to add new timer bases, we have to skip the two
57 * clock ids captured by the cpu-timers. We do this by holding empty
58 * entries rather than doing math adjustment of the clock ids.
59 * This ensures that we capture erroneous accesses to these clock ids
60 * rather than moving them into the range of valid clock id's.
61 */
62DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
63{
64
65 .clock_base =
66 {
67 {
68 .index = CLOCK_REALTIME,
69 .get_time = &ktime_get_real,
70 .resolution = KTIME_LOW_RES,
71 },
72 {
73 .index = CLOCK_MONOTONIC,
74 .get_time = &ktime_get,
75 .resolution = KTIME_LOW_RES,
76 },
77 }
78};
79
80/*
81 * Get the coarse grained time at the softirq based on xtime and
82 * wall_to_monotonic.
83 */
84static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base)
85{
86 ktime_t xtim, tomono;
87 struct timespec xts, tom;
88 unsigned long seq;
89
90 do {
91 seq = read_seqbegin(&xtime_lock);
92 xts = current_kernel_time();
93 tom = wall_to_monotonic;
94 } while (read_seqretry(&xtime_lock, seq));
95
96 xtim = timespec_to_ktime(xts);
97 tomono = timespec_to_ktime(tom);
98 base->clock_base[CLOCK_REALTIME].softirq_time = xtim;
99 base->clock_base[CLOCK_MONOTONIC].softirq_time =
100 ktime_add(xtim, tomono);
101}
102
103/*
104 * Functions and macros which are different for UP/SMP systems are kept in a
105 * single place
106 */
107#ifdef CONFIG_SMP
108
109/*
110 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
111 * means that all timers which are tied to this base via timer->base are
112 * locked, and the base itself is locked too.
113 *
114 * So __run_timers/migrate_timers can safely modify all timers which could
115 * be found on the lists/queues.
116 *
117 * When the timer's base is locked, and the timer removed from list, it is
118 * possible to set timer->base = NULL and drop the lock: the timer remains
119 * locked.
120 */
121static
122struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
123 unsigned long *flags)
124{
125 struct hrtimer_clock_base *base;
126
127 for (;;) {
128 base = timer->base;
129 if (likely(base != NULL)) {
130 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
131 if (likely(base == timer->base))
132 return base;
133 /* The timer has migrated to another CPU: */
134 raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
135 }
136 cpu_relax();
137 }
138}
139
140
141/*
142 * Get the preferred target CPU for NOHZ
143 */
144static int hrtimer_get_target(int this_cpu, int pinned)
145{
146#ifdef CONFIG_NO_HZ
147 if (!pinned && get_sysctl_timer_migration() && idle_cpu(this_cpu)) {
148 int preferred_cpu = get_nohz_load_balancer();
149
150 if (preferred_cpu >= 0)
151 return preferred_cpu;
152 }
153#endif
154 return this_cpu;
155}
156
157/*
158 * With HIGHRES=y we do not migrate the timer when it is expiring
159 * before the next event on the target cpu because we cannot reprogram
160 * the target cpu hardware and we would cause it to fire late.
161 *
162 * Called with cpu_base->lock of target cpu held.
163 */
164static int
165hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
166{
167#ifdef CONFIG_HIGH_RES_TIMERS
168 ktime_t expires;
169
170 if (!new_base->cpu_base->hres_active)
171 return 0;
172
173 expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
174 return expires.tv64 <= new_base->cpu_base->expires_next.tv64;
175#else
176 return 0;
177#endif
178}
179
180/*
181 * Switch the timer base to the current CPU when possible.
182 */
183static inline struct hrtimer_clock_base *
184switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
185 int pinned)
186{
187 struct hrtimer_clock_base *new_base;
188 struct hrtimer_cpu_base *new_cpu_base;
189 int this_cpu = smp_processor_id();
190 int cpu = hrtimer_get_target(this_cpu, pinned);
191
192again:
193 new_cpu_base = &per_cpu(hrtimer_bases, cpu);
194 new_base = &new_cpu_base->clock_base[base->index];
195
196 if (base != new_base) {
197 /*
198 * We are trying to move timer to new_base.
199 * However we can't change timer's base while it is running,
200 * so we keep it on the same CPU. No hassle vs. reprogramming
201 * the event source in the high resolution case. The softirq
202 * code will take care of this when the timer function has
203 * completed. There is no conflict as we hold the lock until
204 * the timer is enqueued.
205 */
206 if (unlikely(hrtimer_callback_running(timer)))
207 return base;
208
209 /* See the comment in lock_timer_base() */
210 timer->base = NULL;
211 raw_spin_unlock(&base->cpu_base->lock);
212 raw_spin_lock(&new_base->cpu_base->lock);
213
214 if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) {
215 cpu = this_cpu;
216 raw_spin_unlock(&new_base->cpu_base->lock);
217 raw_spin_lock(&base->cpu_base->lock);
218 timer->base = base;
219 goto again;
220 }
221 timer->base = new_base;
222 }
223 return new_base;
224}
225
226#else /* CONFIG_SMP */
227
228static inline struct hrtimer_clock_base *
229lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
230{
231 struct hrtimer_clock_base *base = timer->base;
232
233 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
234
235 return base;
236}
237
238# define switch_hrtimer_base(t, b, p) (b)
239
240#endif /* !CONFIG_SMP */
241
242/*
243 * Functions for the union type storage format of ktime_t which are
244 * too large for inlining:
245 */
246#if BITS_PER_LONG < 64
247# ifndef CONFIG_KTIME_SCALAR
248/**
249 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
250 * @kt: addend
251 * @nsec: the scalar nsec value to add
252 *
253 * Returns the sum of kt and nsec in ktime_t format
254 */
255ktime_t ktime_add_ns(const ktime_t kt, u64 nsec)
256{
257 ktime_t tmp;
258
259 if (likely(nsec < NSEC_PER_SEC)) {
260 tmp.tv64 = nsec;
261 } else {
262 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
263
264 tmp = ktime_set((long)nsec, rem);
265 }
266
267 return ktime_add(kt, tmp);
268}
269
270EXPORT_SYMBOL_GPL(ktime_add_ns);
271
272/**
273 * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable
274 * @kt: minuend
275 * @nsec: the scalar nsec value to subtract
276 *
277 * Returns the subtraction of @nsec from @kt in ktime_t format
278 */
279ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec)
280{
281 ktime_t tmp;
282
283 if (likely(nsec < NSEC_PER_SEC)) {
284 tmp.tv64 = nsec;
285 } else {
286 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
287
288 tmp = ktime_set((long)nsec, rem);
289 }
290
291 return ktime_sub(kt, tmp);
292}
293
294EXPORT_SYMBOL_GPL(ktime_sub_ns);
295# endif /* !CONFIG_KTIME_SCALAR */
296
297/*
298 * Divide a ktime value by a nanosecond value
299 */
300u64 ktime_divns(const ktime_t kt, s64 div)
301{
302 u64 dclc;
303 int sft = 0;
304
305 dclc = ktime_to_ns(kt);
306 /* Make sure the divisor is less than 2^32: */
307 while (div >> 32) {
308 sft++;
309 div >>= 1;
310 }
311 dclc >>= sft;
312 do_div(dclc, (unsigned long) div);
313
314 return dclc;
315}
316#endif /* BITS_PER_LONG >= 64 */
317
318/*
319 * Add two ktime values and do a safety check for overflow:
320 */
321ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
322{
323 ktime_t res = ktime_add(lhs, rhs);
324
325 /*
326 * We use KTIME_SEC_MAX here, the maximum timeout which we can
327 * return to user space in a timespec:
328 */
329 if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64)
330 res = ktime_set(KTIME_SEC_MAX, 0);
331
332 return res;
333}
334
335EXPORT_SYMBOL_GPL(ktime_add_safe);
336
337#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
338
339static struct debug_obj_descr hrtimer_debug_descr;
340
341/*
342 * fixup_init is called when:
343 * - an active object is initialized
344 */
345static int hrtimer_fixup_init(void *addr, enum debug_obj_state state)
346{
347 struct hrtimer *timer = addr;
348
349 switch (state) {
350 case ODEBUG_STATE_ACTIVE:
351 hrtimer_cancel(timer);
352 debug_object_init(timer, &hrtimer_debug_descr);
353 return 1;
354 default:
355 return 0;
356 }
357}
358
359/*
360 * fixup_activate is called when:
361 * - an active object is activated
362 * - an unknown object is activated (might be a statically initialized object)
363 */
364static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
365{
366 switch (state) {
367
368 case ODEBUG_STATE_NOTAVAILABLE:
369 WARN_ON_ONCE(1);
370 return 0;
371
372 case ODEBUG_STATE_ACTIVE:
373 WARN_ON(1);
374
375 default:
376 return 0;
377 }
378}
379
380/*
381 * fixup_free is called when:
382 * - an active object is freed
383 */
384static int hrtimer_fixup_free(void *addr, enum debug_obj_state state)
385{
386 struct hrtimer *timer = addr;
387
388 switch (state) {
389 case ODEBUG_STATE_ACTIVE:
390 hrtimer_cancel(timer);
391 debug_object_free(timer, &hrtimer_debug_descr);
392 return 1;
393 default:
394 return 0;
395 }
396}
397
398static struct debug_obj_descr hrtimer_debug_descr = {
399 .name = "hrtimer",
400 .fixup_init = hrtimer_fixup_init,
401 .fixup_activate = hrtimer_fixup_activate,
402 .fixup_free = hrtimer_fixup_free,
403};
404
405static inline void debug_hrtimer_init(struct hrtimer *timer)
406{
407 debug_object_init(timer, &hrtimer_debug_descr);
408}
409
410static inline void debug_hrtimer_activate(struct hrtimer *timer)
411{
412 debug_object_activate(timer, &hrtimer_debug_descr);
413}
414
415static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
416{
417 debug_object_deactivate(timer, &hrtimer_debug_descr);
418}
419
420static inline void debug_hrtimer_free(struct hrtimer *timer)
421{
422 debug_object_free(timer, &hrtimer_debug_descr);
423}
424
425static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
426 enum hrtimer_mode mode);
427
428void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
429 enum hrtimer_mode mode)
430{
431 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
432 __hrtimer_init(timer, clock_id, mode);
433}
434EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
435
436void destroy_hrtimer_on_stack(struct hrtimer *timer)
437{
438 debug_object_free(timer, &hrtimer_debug_descr);
439}
440
441#else
442static inline void debug_hrtimer_init(struct hrtimer *timer) { }
443static inline void debug_hrtimer_activate(struct hrtimer *timer) { }
444static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
445#endif
446
447static inline void
448debug_init(struct hrtimer *timer, clockid_t clockid,
449 enum hrtimer_mode mode)
450{
451 debug_hrtimer_init(timer);
452 trace_hrtimer_init(timer, clockid, mode);
453}
454
455static inline void debug_activate(struct hrtimer *timer)
456{
457 debug_hrtimer_activate(timer);
458 trace_hrtimer_start(timer);
459}
460
461static inline void debug_deactivate(struct hrtimer *timer)
462{
463 debug_hrtimer_deactivate(timer);
464 trace_hrtimer_cancel(timer);
465}
466
467/* High resolution timer related functions */
468#ifdef CONFIG_HIGH_RES_TIMERS
469
470/*
471 * High resolution timer enabled ?
472 */
473static int hrtimer_hres_enabled __read_mostly = 1;
474
475/*
476 * Enable / Disable high resolution mode
477 */
478static int __init setup_hrtimer_hres(char *str)
479{
480 if (!strcmp(str, "off"))
481 hrtimer_hres_enabled = 0;
482 else if (!strcmp(str, "on"))
483 hrtimer_hres_enabled = 1;
484 else
485 return 0;
486 return 1;
487}
488
489__setup("highres=", setup_hrtimer_hres);
490
491/*
492 * hrtimer_high_res_enabled - query, if the highres mode is enabled
493 */
494static inline int hrtimer_is_hres_enabled(void)
495{
496 return hrtimer_hres_enabled;
497}
498
499/*
500 * Is the high resolution mode active ?
501 */
502static inline int hrtimer_hres_active(void)
503{
504 return __get_cpu_var(hrtimer_bases).hres_active;
505}
506
507/*
508 * Reprogram the event source with checking both queues for the
509 * next event
510 * Called with interrupts disabled and base->lock held
511 */
512static void
513hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
514{
515 int i;
516 struct hrtimer_clock_base *base = cpu_base->clock_base;
517 ktime_t expires, expires_next;
518
519 expires_next.tv64 = KTIME_MAX;
520
521 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
522 struct hrtimer *timer;
523
524 if (!base->first)
525 continue;
526 timer = rb_entry(base->first, struct hrtimer, node);
527 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
528 /*
529 * clock_was_set() has changed base->offset so the
530 * result might be negative. Fix it up to prevent a
531 * false positive in clockevents_program_event()
532 */
533 if (expires.tv64 < 0)
534 expires.tv64 = 0;
535 if (expires.tv64 < expires_next.tv64)
536 expires_next = expires;
537 }
538
539 if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64)
540 return;
541
542 cpu_base->expires_next.tv64 = expires_next.tv64;
543
544 if (cpu_base->expires_next.tv64 != KTIME_MAX)
545 tick_program_event(cpu_base->expires_next, 1);
546}
547
548/*
549 * Shared reprogramming for clock_realtime and clock_monotonic
550 *
551 * When a timer is enqueued and expires earlier than the already enqueued
552 * timers, we have to check, whether it expires earlier than the timer for
553 * which the clock event device was armed.
554 *
555 * Called with interrupts disabled and base->cpu_base.lock held
556 */
557static int hrtimer_reprogram(struct hrtimer *timer,
558 struct hrtimer_clock_base *base)
559{
560 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
561 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
562 int res;
563
564 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
565
566 /*
567 * When the callback is running, we do not reprogram the clock event
568 * device. The timer callback is either running on a different CPU or
569 * the callback is executed in the hrtimer_interrupt context. The
570 * reprogramming is handled either by the softirq, which called the
571 * callback or at the end of the hrtimer_interrupt.
572 */
573 if (hrtimer_callback_running(timer))
574 return 0;
575
576 /*
577 * CLOCK_REALTIME timer might be requested with an absolute
578 * expiry time which is less than base->offset. Nothing wrong
579 * about that, just avoid to call into the tick code, which
580 * has now objections against negative expiry values.
581 */
582 if (expires.tv64 < 0)
583 return -ETIME;
584
585 if (expires.tv64 >= cpu_base->expires_next.tv64)
586 return 0;
587
588 /*
589 * If a hang was detected in the last timer interrupt then we
590 * do not schedule a timer which is earlier than the expiry
591 * which we enforced in the hang detection. We want the system
592 * to make progress.
593 */
594 if (cpu_base->hang_detected)
595 return 0;
596
597 /*
598 * Clockevents returns -ETIME, when the event was in the past.
599 */
600 res = tick_program_event(expires, 0);
601 if (!IS_ERR_VALUE(res))
602 cpu_base->expires_next = expires;
603 return res;
604}
605
606
607/*
608 * Retrigger next event is called after clock was set
609 *
610 * Called with interrupts disabled via on_each_cpu()
611 */
612static void retrigger_next_event(void *arg)
613{
614 struct hrtimer_cpu_base *base;
615 struct timespec realtime_offset;
616 unsigned long seq;
617
618 if (!hrtimer_hres_active())
619 return;
620
621 do {
622 seq = read_seqbegin(&xtime_lock);
623 set_normalized_timespec(&realtime_offset,
624 -wall_to_monotonic.tv_sec,
625 -wall_to_monotonic.tv_nsec);
626 } while (read_seqretry(&xtime_lock, seq));
627
628 base = &__get_cpu_var(hrtimer_bases);
629
630 /* Adjust CLOCK_REALTIME offset */
631 raw_spin_lock(&base->lock);
632 base->clock_base[CLOCK_REALTIME].offset =
633 timespec_to_ktime(realtime_offset);
634
635 hrtimer_force_reprogram(base, 0);
636 raw_spin_unlock(&base->lock);
637}
638
639/*
640 * Clock realtime was set
641 *
642 * Change the offset of the realtime clock vs. the monotonic
643 * clock.
644 *
645 * We might have to reprogram the high resolution timer interrupt. On
646 * SMP we call the architecture specific code to retrigger _all_ high
647 * resolution timer interrupts. On UP we just disable interrupts and
648 * call the high resolution interrupt code.
649 */
650void clock_was_set(void)
651{
652 /* Retrigger the CPU local events everywhere */
653 on_each_cpu(retrigger_next_event, NULL, 1);
654}
655
656/*
657 * During resume we might have to reprogram the high resolution timer
658 * interrupt (on the local CPU):
659 */
660void hres_timers_resume(void)
661{
662 WARN_ONCE(!irqs_disabled(),
663 KERN_INFO "hres_timers_resume() called with IRQs enabled!");
664
665 retrigger_next_event(NULL);
666}
667
668/*
669 * Initialize the high resolution related parts of cpu_base
670 */
671static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
672{
673 base->expires_next.tv64 = KTIME_MAX;
674 base->hres_active = 0;
675}
676
677/*
678 * Initialize the high resolution related parts of a hrtimer
679 */
680static inline void hrtimer_init_timer_hres(struct hrtimer *timer)
681{
682}
683
684
685/*
686 * When High resolution timers are active, try to reprogram. Note, that in case
687 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
688 * check happens. The timer gets enqueued into the rbtree. The reprogramming
689 * and expiry check is done in the hrtimer_interrupt or in the softirq.
690 */
691static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
692 struct hrtimer_clock_base *base,
693 int wakeup)
694{
695 if (base->cpu_base->hres_active && hrtimer_reprogram(timer, base)) {
696 if (wakeup) {
697 raw_spin_unlock(&base->cpu_base->lock);
698 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
699 raw_spin_lock(&base->cpu_base->lock);
700 } else
701 __raise_softirq_irqoff(HRTIMER_SOFTIRQ);
702
703 return 1;
704 }
705
706 return 0;
707}
708
709/*
710 * Switch to high resolution mode
711 */
712static int hrtimer_switch_to_hres(void)
713{
714 int cpu = smp_processor_id();
715 struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu);
716 unsigned long flags;
717
718 if (base->hres_active)
719 return 1;
720
721 local_irq_save(flags);
722
723 if (tick_init_highres()) {
724 local_irq_restore(flags);
725 printk(KERN_WARNING "Could not switch to high resolution "
726 "mode on CPU %d\n", cpu);
727 return 0;
728 }
729 base->hres_active = 1;
730 base->clock_base[CLOCK_REALTIME].resolution = KTIME_HIGH_RES;
731 base->clock_base[CLOCK_MONOTONIC].resolution = KTIME_HIGH_RES;
732
733 tick_setup_sched_timer();
734
735 /* "Retrigger" the interrupt to get things going */
736 retrigger_next_event(NULL);
737 local_irq_restore(flags);
738 return 1;
739}
740
741#else
742
743static inline int hrtimer_hres_active(void) { return 0; }
744static inline int hrtimer_is_hres_enabled(void) { return 0; }
745static inline int hrtimer_switch_to_hres(void) { return 0; }
746static inline void
747hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { }
748static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
749 struct hrtimer_clock_base *base,
750 int wakeup)
751{
752 return 0;
753}
754static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
755static inline void hrtimer_init_timer_hres(struct hrtimer *timer) { }
756
757#endif /* CONFIG_HIGH_RES_TIMERS */
758
759static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer)
760{
761#ifdef CONFIG_TIMER_STATS
762 if (timer->start_site)
763 return;
764 timer->start_site = __builtin_return_address(0);
765 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
766 timer->start_pid = current->pid;
767#endif
768}
769
770static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer)
771{
772#ifdef CONFIG_TIMER_STATS
773 timer->start_site = NULL;
774#endif
775}
776
777static inline void timer_stats_account_hrtimer(struct hrtimer *timer)
778{
779#ifdef CONFIG_TIMER_STATS
780 if (likely(!timer_stats_active))
781 return;
782 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
783 timer->function, timer->start_comm, 0);
784#endif
785}
786
787/*
788 * Counterpart to lock_hrtimer_base above:
789 */
790static inline
791void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
792{
793 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
794}
795
796/**
797 * hrtimer_forward - forward the timer expiry
798 * @timer: hrtimer to forward
799 * @now: forward past this time
800 * @interval: the interval to forward
801 *
802 * Forward the timer expiry so it will expire in the future.
803 * Returns the number of overruns.
804 */
805u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
806{
807 u64 orun = 1;
808 ktime_t delta;
809
810 delta = ktime_sub(now, hrtimer_get_expires(timer));
811
812 if (delta.tv64 < 0)
813 return 0;
814
815 if (interval.tv64 < timer->base->resolution.tv64)
816 interval.tv64 = timer->base->resolution.tv64;
817
818 if (unlikely(delta.tv64 >= interval.tv64)) {
819 s64 incr = ktime_to_ns(interval);
820
821 orun = ktime_divns(delta, incr);
822 hrtimer_add_expires_ns(timer, incr * orun);
823 if (hrtimer_get_expires_tv64(timer) > now.tv64)
824 return orun;
825 /*
826 * This (and the ktime_add() below) is the
827 * correction for exact:
828 */
829 orun++;
830 }
831 hrtimer_add_expires(timer, interval);
832
833 return orun;
834}
835EXPORT_SYMBOL_GPL(hrtimer_forward);
836
837/*
838 * enqueue_hrtimer - internal function to (re)start a timer
839 *
840 * The timer is inserted in expiry order. Insertion into the
841 * red black tree is O(log(n)). Must hold the base lock.
842 *
843 * Returns 1 when the new timer is the leftmost timer in the tree.
844 */
845static int enqueue_hrtimer(struct hrtimer *timer,
846 struct hrtimer_clock_base *base)
847{
848 struct rb_node **link = &base->active.rb_node;
849 struct rb_node *parent = NULL;
850 struct hrtimer *entry;
851 int leftmost = 1;
852
853 debug_activate(timer);
854
855 /*
856 * Find the right place in the rbtree:
857 */
858 while (*link) {
859 parent = *link;
860 entry = rb_entry(parent, struct hrtimer, node);
861 /*
862 * We dont care about collisions. Nodes with
863 * the same expiry time stay together.
864 */
865 if (hrtimer_get_expires_tv64(timer) <
866 hrtimer_get_expires_tv64(entry)) {
867 link = &(*link)->rb_left;
868 } else {
869 link = &(*link)->rb_right;
870 leftmost = 0;
871 }
872 }
873
874 /*
875 * Insert the timer to the rbtree and check whether it
876 * replaces the first pending timer
877 */
878 if (leftmost)
879 base->first = &timer->node;
880
881 rb_link_node(&timer->node, parent, link);
882 rb_insert_color(&timer->node, &base->active);
883 /*
884 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
885 * state of a possibly running callback.
886 */
887 timer->state |= HRTIMER_STATE_ENQUEUED;
888
889 return leftmost;
890}
891
892/*
893 * __remove_hrtimer - internal function to remove a timer
894 *
895 * Caller must hold the base lock.
896 *
897 * High resolution timer mode reprograms the clock event device when the
898 * timer is the one which expires next. The caller can disable this by setting
899 * reprogram to zero. This is useful, when the context does a reprogramming
900 * anyway (e.g. timer interrupt)
901 */
902static void __remove_hrtimer(struct hrtimer *timer,
903 struct hrtimer_clock_base *base,
904 unsigned long newstate, int reprogram)
905{
906 if (!(timer->state & HRTIMER_STATE_ENQUEUED))
907 goto out;
908
909 /*
910 * Remove the timer from the rbtree and replace the first
911 * entry pointer if necessary.
912 */
913 if (base->first == &timer->node) {
914 base->first = rb_next(&timer->node);
915#ifdef CONFIG_HIGH_RES_TIMERS
916 /* Reprogram the clock event device. if enabled */
917 if (reprogram && hrtimer_hres_active()) {
918 ktime_t expires;
919
920 expires = ktime_sub(hrtimer_get_expires(timer),
921 base->offset);
922 if (base->cpu_base->expires_next.tv64 == expires.tv64)
923 hrtimer_force_reprogram(base->cpu_base, 1);
924 }
925#endif
926 }
927 rb_erase(&timer->node, &base->active);
928out:
929 timer->state = newstate;
930}
931
932/*
933 * remove hrtimer, called with base lock held
934 */
935static inline int
936remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
937{
938 if (hrtimer_is_queued(timer)) {
939 int reprogram;
940
941 /*
942 * Remove the timer and force reprogramming when high
943 * resolution mode is active and the timer is on the current
944 * CPU. If we remove a timer on another CPU, reprogramming is
945 * skipped. The interrupt event on this CPU is fired and
946 * reprogramming happens in the interrupt handler. This is a
947 * rare case and less expensive than a smp call.
948 */
949 debug_deactivate(timer);
950 timer_stats_hrtimer_clear_start_info(timer);
951 reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases);
952 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE,
953 reprogram);
954 return 1;
955 }
956 return 0;
957}
958
959int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
960 unsigned long delta_ns, const enum hrtimer_mode mode,
961 int wakeup)
962{
963 struct hrtimer_clock_base *base, *new_base;
964 unsigned long flags;
965 int ret, leftmost;
966
967 base = lock_hrtimer_base(timer, &flags);
968
969 /* Remove an active timer from the queue: */
970 ret = remove_hrtimer(timer, base);
971
972 /* Switch the timer base, if necessary: */
973 new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
974
975 if (mode & HRTIMER_MODE_REL) {
976 tim = ktime_add_safe(tim, new_base->get_time());
977 /*
978 * CONFIG_TIME_LOW_RES is a temporary way for architectures
979 * to signal that they simply return xtime in
980 * do_gettimeoffset(). In this case we want to round up by
981 * resolution when starting a relative timer, to avoid short
982 * timeouts. This will go away with the GTOD framework.
983 */
984#ifdef CONFIG_TIME_LOW_RES
985 tim = ktime_add_safe(tim, base->resolution);
986#endif
987 }
988
989 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
990
991 timer_stats_hrtimer_set_start_info(timer);
992
993 leftmost = enqueue_hrtimer(timer, new_base);
994
995 /*
996 * Only allow reprogramming if the new base is on this CPU.
997 * (it might still be on another CPU if the timer was pending)
998 *
999 * XXX send_remote_softirq() ?
1000 */
1001 if (leftmost && new_base->cpu_base == &__get_cpu_var(hrtimer_bases))
1002 hrtimer_enqueue_reprogram(timer, new_base, wakeup);
1003
1004 unlock_hrtimer_base(timer, &flags);
1005
1006 return ret;
1007}
1008
1009/**
1010 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
1011 * @timer: the timer to be added
1012 * @tim: expiry time
1013 * @delta_ns: "slack" range for the timer
1014 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
1015 *
1016 * Returns:
1017 * 0 on success
1018 * 1 when the timer was active
1019 */
1020int hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1021 unsigned long delta_ns, const enum hrtimer_mode mode)
1022{
1023 return __hrtimer_start_range_ns(timer, tim, delta_ns, mode, 1);
1024}
1025EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1026
1027/**
1028 * hrtimer_start - (re)start an hrtimer on the current CPU
1029 * @timer: the timer to be added
1030 * @tim: expiry time
1031 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
1032 *
1033 * Returns:
1034 * 0 on success
1035 * 1 when the timer was active
1036 */
1037int
1038hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
1039{
1040 return __hrtimer_start_range_ns(timer, tim, 0, mode, 1);
1041}
1042EXPORT_SYMBOL_GPL(hrtimer_start);
1043
1044
1045/**
1046 * hrtimer_try_to_cancel - try to deactivate a timer
1047 * @timer: hrtimer to stop
1048 *
1049 * Returns:
1050 * 0 when the timer was not active
1051 * 1 when the timer was active
1052 * -1 when the timer is currently excuting the callback function and
1053 * cannot be stopped
1054 */
1055int hrtimer_try_to_cancel(struct hrtimer *timer)
1056{
1057 struct hrtimer_clock_base *base;
1058 unsigned long flags;
1059 int ret = -1;
1060
1061 base = lock_hrtimer_base(timer, &flags);
1062
1063 if (!hrtimer_callback_running(timer))
1064 ret = remove_hrtimer(timer, base);
1065
1066 unlock_hrtimer_base(timer, &flags);
1067
1068 return ret;
1069
1070}
1071EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1072
1073/**
1074 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1075 * @timer: the timer to be cancelled
1076 *
1077 * Returns:
1078 * 0 when the timer was not active
1079 * 1 when the timer was active
1080 */
1081int hrtimer_cancel(struct hrtimer *timer)
1082{
1083 for (;;) {
1084 int ret = hrtimer_try_to_cancel(timer);
1085
1086 if (ret >= 0)
1087 return ret;
1088 cpu_relax();
1089 }
1090}
1091EXPORT_SYMBOL_GPL(hrtimer_cancel);
1092
1093/**
1094 * hrtimer_get_remaining - get remaining time for the timer
1095 * @timer: the timer to read
1096 */
1097ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
1098{
1099 struct hrtimer_clock_base *base;
1100 unsigned long flags;
1101 ktime_t rem;
1102
1103 base = lock_hrtimer_base(timer, &flags);
1104 rem = hrtimer_expires_remaining(timer);
1105 unlock_hrtimer_base(timer, &flags);
1106
1107 return rem;
1108}
1109EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
1110
1111#ifdef CONFIG_NO_HZ
1112/**
1113 * hrtimer_get_next_event - get the time until next expiry event
1114 *
1115 * Returns the delta to the next expiry event or KTIME_MAX if no timer
1116 * is pending.
1117 */
1118ktime_t hrtimer_get_next_event(void)
1119{
1120 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1121 struct hrtimer_clock_base *base = cpu_base->clock_base;
1122 ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
1123 unsigned long flags;
1124 int i;
1125
1126 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1127
1128 if (!hrtimer_hres_active()) {
1129 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
1130 struct hrtimer *timer;
1131
1132 if (!base->first)
1133 continue;
1134
1135 timer = rb_entry(base->first, struct hrtimer, node);
1136 delta.tv64 = hrtimer_get_expires_tv64(timer);
1137 delta = ktime_sub(delta, base->get_time());
1138 if (delta.tv64 < mindelta.tv64)
1139 mindelta.tv64 = delta.tv64;
1140 }
1141 }
1142
1143 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1144
1145 if (mindelta.tv64 < 0)
1146 mindelta.tv64 = 0;
1147 return mindelta;
1148}
1149#endif
1150
1151static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1152 enum hrtimer_mode mode)
1153{
1154 struct hrtimer_cpu_base *cpu_base;
1155
1156 memset(timer, 0, sizeof(struct hrtimer));
1157
1158 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1159
1160 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
1161 clock_id = CLOCK_MONOTONIC;
1162
1163 timer->base = &cpu_base->clock_base[clock_id];
1164 hrtimer_init_timer_hres(timer);
1165
1166#ifdef CONFIG_TIMER_STATS
1167 timer->start_site = NULL;
1168 timer->start_pid = -1;
1169 memset(timer->start_comm, 0, TASK_COMM_LEN);
1170#endif
1171}
1172
1173/**
1174 * hrtimer_init - initialize a timer to the given clock
1175 * @timer: the timer to be initialized
1176 * @clock_id: the clock to be used
1177 * @mode: timer mode abs/rel
1178 */
1179void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1180 enum hrtimer_mode mode)
1181{
1182 debug_init(timer, clock_id, mode);
1183 __hrtimer_init(timer, clock_id, mode);
1184}
1185EXPORT_SYMBOL_GPL(hrtimer_init);
1186
1187/**
1188 * hrtimer_get_res - get the timer resolution for a clock
1189 * @which_clock: which clock to query
1190 * @tp: pointer to timespec variable to store the resolution
1191 *
1192 * Store the resolution of the clock selected by @which_clock in the
1193 * variable pointed to by @tp.
1194 */
1195int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
1196{
1197 struct hrtimer_cpu_base *cpu_base;
1198
1199 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1200 *tp = ktime_to_timespec(cpu_base->clock_base[which_clock].resolution);
1201
1202 return 0;
1203}
1204EXPORT_SYMBOL_GPL(hrtimer_get_res);
1205
1206static void __run_hrtimer(struct hrtimer *timer, ktime_t *now)
1207{
1208 struct hrtimer_clock_base *base = timer->base;
1209 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
1210 enum hrtimer_restart (*fn)(struct hrtimer *);
1211 int restart;
1212
1213 WARN_ON(!irqs_disabled());
1214
1215 debug_deactivate(timer);
1216 __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
1217 timer_stats_account_hrtimer(timer);
1218 fn = timer->function;
1219
1220 /*
1221 * Because we run timers from hardirq context, there is no chance
1222 * they get migrated to another cpu, therefore its safe to unlock
1223 * the timer base.
1224 */
1225 raw_spin_unlock(&cpu_base->lock);
1226 trace_hrtimer_expire_entry(timer, now);
1227 restart = fn(timer);
1228 trace_hrtimer_expire_exit(timer);
1229 raw_spin_lock(&cpu_base->lock);
1230
1231 /*
1232 * Note: We clear the CALLBACK bit after enqueue_hrtimer and
1233 * we do not reprogramm the event hardware. Happens either in
1234 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1235 */
1236 if (restart != HRTIMER_NORESTART) {
1237 BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
1238 enqueue_hrtimer(timer, base);
1239 }
1240 timer->state &= ~HRTIMER_STATE_CALLBACK;
1241}
1242
1243#ifdef CONFIG_HIGH_RES_TIMERS
1244
1245/*
1246 * High resolution timer interrupt
1247 * Called with interrupts disabled
1248 */
1249void hrtimer_interrupt(struct clock_event_device *dev)
1250{
1251 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1252 struct hrtimer_clock_base *base;
1253 ktime_t expires_next, now, entry_time, delta;
1254 int i, retries = 0;
1255
1256 BUG_ON(!cpu_base->hres_active);
1257 cpu_base->nr_events++;
1258 dev->next_event.tv64 = KTIME_MAX;
1259
1260 entry_time = now = ktime_get();
1261retry:
1262 expires_next.tv64 = KTIME_MAX;
1263
1264 raw_spin_lock(&cpu_base->lock);
1265 /*
1266 * We set expires_next to KTIME_MAX here with cpu_base->lock
1267 * held to prevent that a timer is enqueued in our queue via
1268 * the migration code. This does not affect enqueueing of
1269 * timers which run their callback and need to be requeued on
1270 * this CPU.
1271 */
1272 cpu_base->expires_next.tv64 = KTIME_MAX;
1273
1274 base = cpu_base->clock_base;
1275
1276 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1277 ktime_t basenow;
1278 struct rb_node *node;
1279
1280 basenow = ktime_add(now, base->offset);
1281
1282 while ((node = base->first)) {
1283 struct hrtimer *timer;
1284
1285 timer = rb_entry(node, struct hrtimer, node);
1286
1287 /*
1288 * The immediate goal for using the softexpires is
1289 * minimizing wakeups, not running timers at the
1290 * earliest interrupt after their soft expiration.
1291 * This allows us to avoid using a Priority Search
1292 * Tree, which can answer a stabbing querry for
1293 * overlapping intervals and instead use the simple
1294 * BST we already have.
1295 * We don't add extra wakeups by delaying timers that
1296 * are right-of a not yet expired timer, because that
1297 * timer will have to trigger a wakeup anyway.
1298 */
1299
1300 if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) {
1301 ktime_t expires;
1302
1303 expires = ktime_sub(hrtimer_get_expires(timer),
1304 base->offset);
1305 if (expires.tv64 < expires_next.tv64)
1306 expires_next = expires;
1307 break;
1308 }
1309
1310 __run_hrtimer(timer, &basenow);
1311 }
1312 base++;
1313 }
1314
1315 /*
1316 * Store the new expiry value so the migration code can verify
1317 * against it.
1318 */
1319 cpu_base->expires_next = expires_next;
1320 raw_spin_unlock(&cpu_base->lock);
1321
1322 /* Reprogramming necessary ? */
1323 if (expires_next.tv64 == KTIME_MAX ||
1324 !tick_program_event(expires_next, 0)) {
1325 cpu_base->hang_detected = 0;
1326 return;
1327 }
1328
1329 /*
1330 * The next timer was already expired due to:
1331 * - tracing
1332 * - long lasting callbacks
1333 * - being scheduled away when running in a VM
1334 *
1335 * We need to prevent that we loop forever in the hrtimer
1336 * interrupt routine. We give it 3 attempts to avoid
1337 * overreacting on some spurious event.
1338 */
1339 now = ktime_get();
1340 cpu_base->nr_retries++;
1341 if (++retries < 3)
1342 goto retry;
1343 /*
1344 * Give the system a chance to do something else than looping
1345 * here. We stored the entry time, so we know exactly how long
1346 * we spent here. We schedule the next event this amount of
1347 * time away.
1348 */
1349 cpu_base->nr_hangs++;
1350 cpu_base->hang_detected = 1;
1351 delta = ktime_sub(now, entry_time);
1352 if (delta.tv64 > cpu_base->max_hang_time.tv64)
1353 cpu_base->max_hang_time = delta;
1354 /*
1355 * Limit it to a sensible value as we enforce a longer
1356 * delay. Give the CPU at least 100ms to catch up.
1357 */
1358 if (delta.tv64 > 100 * NSEC_PER_MSEC)
1359 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1360 else
1361 expires_next = ktime_add(now, delta);
1362 tick_program_event(expires_next, 1);
1363 printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n",
1364 ktime_to_ns(delta));
1365}
1366
1367/*
1368 * local version of hrtimer_peek_ahead_timers() called with interrupts
1369 * disabled.
1370 */
1371static void __hrtimer_peek_ahead_timers(void)
1372{
1373 struct tick_device *td;
1374
1375 if (!hrtimer_hres_active())
1376 return;
1377
1378 td = &__get_cpu_var(tick_cpu_device);
1379 if (td && td->evtdev)
1380 hrtimer_interrupt(td->evtdev);
1381}
1382
1383/**
1384 * hrtimer_peek_ahead_timers -- run soft-expired timers now
1385 *
1386 * hrtimer_peek_ahead_timers will peek at the timer queue of
1387 * the current cpu and check if there are any timers for which
1388 * the soft expires time has passed. If any such timers exist,
1389 * they are run immediately and then removed from the timer queue.
1390 *
1391 */
1392void hrtimer_peek_ahead_timers(void)
1393{
1394 unsigned long flags;
1395
1396 local_irq_save(flags);
1397 __hrtimer_peek_ahead_timers();
1398 local_irq_restore(flags);
1399}
1400
1401static void run_hrtimer_softirq(struct softirq_action *h)
1402{
1403 hrtimer_peek_ahead_timers();
1404}
1405
1406#else /* CONFIG_HIGH_RES_TIMERS */
1407
1408static inline void __hrtimer_peek_ahead_timers(void) { }
1409
1410#endif /* !CONFIG_HIGH_RES_TIMERS */
1411
1412/*
1413 * Called from timer softirq every jiffy, expire hrtimers:
1414 *
1415 * For HRT its the fall back code to run the softirq in the timer
1416 * softirq context in case the hrtimer initialization failed or has
1417 * not been done yet.
1418 */
1419void hrtimer_run_pending(void)
1420{
1421 if (hrtimer_hres_active())
1422 return;
1423
1424 /*
1425 * This _is_ ugly: We have to check in the softirq context,
1426 * whether we can switch to highres and / or nohz mode. The
1427 * clocksource switch happens in the timer interrupt with
1428 * xtime_lock held. Notification from there only sets the
1429 * check bit in the tick_oneshot code, otherwise we might
1430 * deadlock vs. xtime_lock.
1431 */
1432 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1433 hrtimer_switch_to_hres();
1434}
1435
1436/*
1437 * Called from hardirq context every jiffy
1438 */
1439void hrtimer_run_queues(void)
1440{
1441 struct rb_node *node;
1442 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1443 struct hrtimer_clock_base *base;
1444 int index, gettime = 1;
1445
1446 if (hrtimer_hres_active())
1447 return;
1448
1449 for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) {
1450 base = &cpu_base->clock_base[index];
1451
1452 if (!base->first)
1453 continue;
1454
1455 if (gettime) {
1456 hrtimer_get_softirq_time(cpu_base);
1457 gettime = 0;
1458 }
1459
1460 raw_spin_lock(&cpu_base->lock);
1461
1462 while ((node = base->first)) {
1463 struct hrtimer *timer;
1464
1465 timer = rb_entry(node, struct hrtimer, node);
1466 if (base->softirq_time.tv64 <=
1467 hrtimer_get_expires_tv64(timer))
1468 break;
1469
1470 __run_hrtimer(timer, &base->softirq_time);
1471 }
1472 raw_spin_unlock(&cpu_base->lock);
1473 }
1474}
1475
1476/*
1477 * Sleep related functions:
1478 */
1479static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1480{
1481 struct hrtimer_sleeper *t =
1482 container_of(timer, struct hrtimer_sleeper, timer);
1483 struct task_struct *task = t->task;
1484
1485 t->task = NULL;
1486 if (task)
1487 wake_up_process(task);
1488
1489 return HRTIMER_NORESTART;
1490}
1491
1492void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1493{
1494 sl->timer.function = hrtimer_wakeup;
1495 sl->task = task;
1496}
1497EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1498
1499static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1500{
1501 hrtimer_init_sleeper(t, current);
1502
1503 do {
1504 set_current_state(TASK_INTERRUPTIBLE);
1505 hrtimer_start_expires(&t->timer, mode);
1506 if (!hrtimer_active(&t->timer))
1507 t->task = NULL;
1508
1509 if (likely(t->task))
1510 schedule();
1511
1512 hrtimer_cancel(&t->timer);
1513 mode = HRTIMER_MODE_ABS;
1514
1515 } while (t->task && !signal_pending(current));
1516
1517 __set_current_state(TASK_RUNNING);
1518
1519 return t->task == NULL;
1520}
1521
1522static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp)
1523{
1524 struct timespec rmt;
1525 ktime_t rem;
1526
1527 rem = hrtimer_expires_remaining(timer);
1528 if (rem.tv64 <= 0)
1529 return 0;
1530 rmt = ktime_to_timespec(rem);
1531
1532 if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
1533 return -EFAULT;
1534
1535 return 1;
1536}
1537
1538long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1539{
1540 struct hrtimer_sleeper t;
1541 struct timespec __user *rmtp;
1542 int ret = 0;
1543
1544 hrtimer_init_on_stack(&t.timer, restart->nanosleep.index,
1545 HRTIMER_MODE_ABS);
1546 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1547
1548 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1549 goto out;
1550
1551 rmtp = restart->nanosleep.rmtp;
1552 if (rmtp) {
1553 ret = update_rmtp(&t.timer, rmtp);
1554 if (ret <= 0)
1555 goto out;
1556 }
1557
1558 /* The other values in restart are already filled in */
1559 ret = -ERESTART_RESTARTBLOCK;
1560out:
1561 destroy_hrtimer_on_stack(&t.timer);
1562 return ret;
1563}
1564
1565long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1566 const enum hrtimer_mode mode, const clockid_t clockid)
1567{
1568 struct restart_block *restart;
1569 struct hrtimer_sleeper t;
1570 int ret = 0;
1571 unsigned long slack;
1572
1573 slack = current->timer_slack_ns;
1574 if (rt_task(current))
1575 slack = 0;
1576
1577 hrtimer_init_on_stack(&t.timer, clockid, mode);
1578 hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
1579 if (do_nanosleep(&t, mode))
1580 goto out;
1581
1582 /* Absolute timers do not update the rmtp value and restart: */
1583 if (mode == HRTIMER_MODE_ABS) {
1584 ret = -ERESTARTNOHAND;
1585 goto out;
1586 }
1587
1588 if (rmtp) {
1589 ret = update_rmtp(&t.timer, rmtp);
1590 if (ret <= 0)
1591 goto out;
1592 }
1593
1594 restart = &current_thread_info()->restart_block;
1595 restart->fn = hrtimer_nanosleep_restart;
1596 restart->nanosleep.index = t.timer.base->index;
1597 restart->nanosleep.rmtp = rmtp;
1598 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1599
1600 ret = -ERESTART_RESTARTBLOCK;
1601out:
1602 destroy_hrtimer_on_stack(&t.timer);
1603 return ret;
1604}
1605
1606SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1607 struct timespec __user *, rmtp)
1608{
1609 struct timespec tu;
1610
1611 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1612 return -EFAULT;
1613
1614 if (!timespec_valid(&tu))
1615 return -EINVAL;
1616
1617 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1618}
1619
1620/*
1621 * Functions related to boot-time initialization:
1622 */
1623static void __cpuinit init_hrtimers_cpu(int cpu)
1624{
1625 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1626 int i;
1627
1628 raw_spin_lock_init(&cpu_base->lock);
1629
1630 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
1631 cpu_base->clock_base[i].cpu_base = cpu_base;
1632
1633 hrtimer_init_hres(cpu_base);
1634}
1635
1636#ifdef CONFIG_HOTPLUG_CPU
1637
1638static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1639 struct hrtimer_clock_base *new_base)
1640{
1641 struct hrtimer *timer;
1642 struct rb_node *node;
1643
1644 while ((node = rb_first(&old_base->active))) {
1645 timer = rb_entry(node, struct hrtimer, node);
1646 BUG_ON(hrtimer_callback_running(timer));
1647 debug_deactivate(timer);
1648
1649 /*
1650 * Mark it as STATE_MIGRATE not INACTIVE otherwise the
1651 * timer could be seen as !active and just vanish away
1652 * under us on another CPU
1653 */
1654 __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0);
1655 timer->base = new_base;
1656 /*
1657 * Enqueue the timers on the new cpu. This does not
1658 * reprogram the event device in case the timer
1659 * expires before the earliest on this CPU, but we run
1660 * hrtimer_interrupt after we migrated everything to
1661 * sort out already expired timers and reprogram the
1662 * event device.
1663 */
1664 enqueue_hrtimer(timer, new_base);
1665
1666 /* Clear the migration state bit */
1667 timer->state &= ~HRTIMER_STATE_MIGRATE;
1668 }
1669}
1670
1671static void migrate_hrtimers(int scpu)
1672{
1673 struct hrtimer_cpu_base *old_base, *new_base;
1674 int i;
1675
1676 BUG_ON(cpu_online(scpu));
1677 tick_cancel_sched_timer(scpu);
1678
1679 local_irq_disable();
1680 old_base = &per_cpu(hrtimer_bases, scpu);
1681 new_base = &__get_cpu_var(hrtimer_bases);
1682 /*
1683 * The caller is globally serialized and nobody else
1684 * takes two locks at once, deadlock is not possible.
1685 */
1686 raw_spin_lock(&new_base->lock);
1687 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1688
1689 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1690 migrate_hrtimer_list(&old_base->clock_base[i],
1691 &new_base->clock_base[i]);
1692 }
1693
1694 raw_spin_unlock(&old_base->lock);
1695 raw_spin_unlock(&new_base->lock);
1696
1697 /* Check, if we got expired work to do */
1698 __hrtimer_peek_ahead_timers();
1699 local_irq_enable();
1700}
1701
1702#endif /* CONFIG_HOTPLUG_CPU */
1703
1704static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self,
1705 unsigned long action, void *hcpu)
1706{
1707 int scpu = (long)hcpu;
1708
1709 switch (action) {
1710
1711 case CPU_UP_PREPARE:
1712 case CPU_UP_PREPARE_FROZEN:
1713 init_hrtimers_cpu(scpu);
1714 break;
1715
1716#ifdef CONFIG_HOTPLUG_CPU
1717 case CPU_DYING:
1718 case CPU_DYING_FROZEN:
1719 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING, &scpu);
1720 break;
1721 case CPU_DEAD:
1722 case CPU_DEAD_FROZEN:
1723 {
1724 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &scpu);
1725 migrate_hrtimers(scpu);
1726 break;
1727 }
1728#endif
1729
1730 default:
1731 break;
1732 }
1733
1734 return NOTIFY_OK;
1735}
1736
1737static struct notifier_block __cpuinitdata hrtimers_nb = {
1738 .notifier_call = hrtimer_cpu_notify,
1739};
1740
1741void __init hrtimers_init(void)
1742{
1743 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1744 (void *)(long)smp_processor_id());
1745 register_cpu_notifier(&hrtimers_nb);
1746#ifdef CONFIG_HIGH_RES_TIMERS
1747 open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq);
1748#endif
1749}
1750
1751/**
1752 * schedule_hrtimeout_range - sleep until timeout
1753 * @expires: timeout value (ktime_t)
1754 * @delta: slack in expires timeout (ktime_t)
1755 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1756 *
1757 * Make the current task sleep until the given expiry time has
1758 * elapsed. The routine will return immediately unless
1759 * the current task state has been set (see set_current_state()).
1760 *
1761 * The @delta argument gives the kernel the freedom to schedule the
1762 * actual wakeup to a time that is both power and performance friendly.
1763 * The kernel give the normal best effort behavior for "@expires+@delta",
1764 * but may decide to fire the timer earlier, but no earlier than @expires.
1765 *
1766 * You can set the task state as follows -
1767 *
1768 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1769 * pass before the routine returns.
1770 *
1771 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1772 * delivered to the current task.
1773 *
1774 * The current task state is guaranteed to be TASK_RUNNING when this
1775 * routine returns.
1776 *
1777 * Returns 0 when the timer has expired otherwise -EINTR
1778 */
1779int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta,
1780 const enum hrtimer_mode mode)
1781{
1782 struct hrtimer_sleeper t;
1783
1784 /*
1785 * Optimize when a zero timeout value is given. It does not
1786 * matter whether this is an absolute or a relative time.
1787 */
1788 if (expires && !expires->tv64) {
1789 __set_current_state(TASK_RUNNING);
1790 return 0;
1791 }
1792
1793 /*
1794 * A NULL parameter means "inifinte"
1795 */
1796 if (!expires) {
1797 schedule();
1798 __set_current_state(TASK_RUNNING);
1799 return -EINTR;
1800 }
1801
1802 hrtimer_init_on_stack(&t.timer, CLOCK_MONOTONIC, mode);
1803 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1804
1805 hrtimer_init_sleeper(&t, current);
1806
1807 hrtimer_start_expires(&t.timer, mode);
1808 if (!hrtimer_active(&t.timer))
1809 t.task = NULL;
1810
1811 if (likely(t.task))
1812 schedule();
1813
1814 hrtimer_cancel(&t.timer);
1815 destroy_hrtimer_on_stack(&t.timer);
1816
1817 __set_current_state(TASK_RUNNING);
1818
1819 return !t.task ? 0 : -EINTR;
1820}
1821EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1822
1823/**
1824 * schedule_hrtimeout - sleep until timeout
1825 * @expires: timeout value (ktime_t)
1826 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1827 *
1828 * Make the current task sleep until the given expiry time has
1829 * elapsed. The routine will return immediately unless
1830 * the current task state has been set (see set_current_state()).
1831 *
1832 * You can set the task state as follows -
1833 *
1834 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1835 * pass before the routine returns.
1836 *
1837 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1838 * delivered to the current task.
1839 *
1840 * The current task state is guaranteed to be TASK_RUNNING when this
1841 * routine returns.
1842 *
1843 * Returns 0 when the timer has expired otherwise -EINTR
1844 */
1845int __sched schedule_hrtimeout(ktime_t *expires,
1846 const enum hrtimer_mode mode)
1847{
1848 return schedule_hrtimeout_range(expires, 0, mode);
1849}
1850EXPORT_SYMBOL_GPL(schedule_hrtimeout);